How to keep depth the same across conv3d layer

[drydenwiebe]

Hello,

I am trying to use a conv3d layer to for images with multiple frames, but I would like the output to have the same depth before and after the layer (ie: D == D_out like this.

Right now my input shape is:

(6, 5, 3, 78, 136)

This is with a batch size of 6, and a depth of 5 (ie: D = 5)

The conv3d block is initialized as:

torch.nn.Conv3d(3, 64, kernel_size=3, padding=1, stride=(1, 2, 2))

When I run this, the error I get is:

RuntimeError: Given groups=1, weight of size [64, 2, 3, 3, 3], expected input[6, 5, 2, 78, 136] to have 2 channels, but got 5 channels instead

I guess what I want is weights of size [64, 5, 3, 3, 3] to have D = D_out. How can I accomplish this?

[drydenwiebe]

According to the formula for D_out here

For this case this is:

Which should be 5, as expected so I am not sure what is happening.

[drydenwiebe]

Figured it out. My channels and depth were swapped for my input tensor.